Fix truncate-toward-zero distortion

[argilo]

Fixes #673.

Most kernels with integer outputs use rintf to round the output. But volk_32f_x2_dot_prod_16i and volk_32fc_s32f_deinterleave_real_16i cast directly to int16_t, which introduces truncate-toward-zero distortion. Here I've switched the affected kernels to use rintf instead.

[argilo]

GNU Radio uses volk_32f_x2_dot_prod_16i in the Float->Short variants of the Decimating & Interpolating FIR Filter blocks. Distortion can be observed with the following flow graph:

Before:

After:

[argilo]

It's a bit easier to see what's going on with the floating point signal in the picture as well:

Before:

After:

[argilo]

It turns out there are two GNU Radio tests which assert the truncate-toward-zero behaviour:

https://github.com/gnuradio/gnuradio/blob/9b22fd38b8259f33928c7b3ff7868e0bc4b60d1e/gr-filter/python/filter/qa_fir_filter.py#L160-L188

Those tests now fail. I'll have to think about what the best solution is.

[argilo]

The failing tests were fixed in https://github.com/gnuradio/gnuradio/pull/6980.